video thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

Crisp Boundary Detection Using Pointwise Mutual Information

Published on Feb 4, 20252430 Views

Detecting boundaries between semantically meaningful objects in visual scenes is an important component of many vision algorithms. In this paper, we propose a novel method for detecting such boundarie

Related categories

Presentation

Crisp Boundary Detection Using Pointwise Mutual Information00:00
Goal00:19
How do you find a boundary?00:46
Key observation - 102:13
Key observation - 202:48
Key observation - 304:31
Is PMI informative about object boundaries? - 105:37
Is PMI informative about object boundaries? - 206:17
PMI leverages internal image statistics - 106:35
PMI leverages internal image statistics - 207:42
Algorithm08:29
Algorithm outline08:33
Step 1: Estimate feature co-occurrence distribution P(A,B)09:01
Step 2: Derive PMI(A,B) from feature co-occurrence distribution10:09
Step 3: Use PMI as affinity between each pair of nearby pixels - 110:17
Step 3: Use PMI as affinity between each pair of nearby pixels - 210:55
Step 4: Group pixels based on affinity (spectral clustering)11:33
Results - 111:44
Results - 211:47
Results - 313:08
Performance on BSDS50013:20
Works on diverse stimuli14:00
Boundary detection in XYT14:41
Summary15:01
Code available15:25